J. PAIS-RIBEIRO1, I. SILVA2, T. FERREIRA3, A. MARTINS3, R. MENESES2, & M. BALTAR1
1Faculdade de Psicologia e de Ciências da Educação, Universidade do Porto, 2Universidade Fernando
Pessoa, Porto, 3Escola Superior de Enfermagem S. João, Porto, Portugal
Abstract
The study aims to develop and assess metric proprieties of the Portuguese version of the Hospital Anxiety and Depression Scale. A sequential sample includes 1322 participants diagnosed with cancer, stroke, epilepsy, coronary heart disease, diabetes, myotonic dystrophy, obstructive sleep apnoea, depression and a non-disease group, which completed the HADS. The first step includes translation, retroversion, inspection for lexical equivalence and content validity, and cognitive debriefing. Then we reproduce oblique exploratory factor analysis and use confirmatory factor analysis. We explore the sensibility of the questionnaire. The validation process of the Portuguese HADS version shows metric properties similar to those in international studies, suggesting that it measures the same constructs, in the same way, as the original HADS form.
Keywords: Anxiety and depression, metric properties, scale adaptation
Introduction
The Hospital Anxiety and Depression Scale (HADS) was designed to aid the clinician in recognizing emotional components of physical illness (Snaith & Zigmond, 1994) as a screening device for anxiety and depression in a general hospital setting and has proved to be useful in the assessment of changes in the patient’s emotional state (Zigmond & Snaith, 1983). These components may increase the distress involved in physical illness, make diagnosing more difficult or prolong the recovery time. Once recognized, emotional distress may respond well to appropriate antidepressant and anxiety management measures.
Zigmond and Snaith (1983) explain that emotional disorder may sometimes result from stress caused by physical disability but somatic symptoms leading to referrals to medical or surgical settings may be a manifestation of emotional states, namely anxiety or depression, with no basis in organic pathology. Because of this, authors selected the HADS items to be relatively unaffected by concurrent physical illness such as loss of appetite and sleep disturbance, which may be symptomatic of either emotional disturbance or physical illness (Snaith, 2003).
Correspondence: J. Pais-Ribeiro, Faculdade de Psicologia e de Cieˆncias da Educac¸a˜o, Universidade do Porto, R. do Campo Alegre, 1021/1055, 4169-004 Porto, Portugal. Tel: (351) 965045590. E-mail: jlpr@fpce.up.pt
ISSN 1354-8506 print/ISSN 1465-3966 online ª 2007 Taylor & Francis DOI: 10.1080/13548500500524088
Traditional emotional distress measures are time and energy consuming and to make cost- effective screening of emotional disorders feasible, development of brief questionnaires like HADS is necessary. To be useful in disease settings the questionnaire must be brief, acceptable to patients, readily comprehended and provide clear information on the interpretation of scores.
Snaith and Zigmond (1994) explained that HADS is a present-state instrument with emphasis on the condition within the preceding few days; subscales were established for the purpose of screening for the presence of mood as a disorder state. Despite the term ‘‘hospital’’ in the title, subsequent work has shown the scale to be valid in primary care and community settings. It is also useful in general psychiatric and clinical psychological work.
Some researchers defend that ‘‘the HADS score does not accurately identify the presence of major depressive disorder in either medical or psychiatric patients, and should not therefore be used in research studies for this purpose’’ (Silverstone, 1994, p. 449). Silverstone states that HADS remains clinically useful for its original screening purpose. Further validation studies of the English and of foreign language translations of the HADS were undertaken in a variety of settings and centres. In a recent review, Bjelland, Dahl, Haug, and Neckelmann (2002) concluded: ‘‘The HADS was found to perform well in assessing severity and caseness of anxiety disorders and depression in both somatic, psychiatric in primary care patients and in the general population’’ (p. 69). In spite of to be developed to identify two separate constructs (anxiety and depression), Herrmann (1997) and Johnston, Pollard and Hennessy (2000) propose that a unique score based on the 14 items in HADS can be used to measure emotional distress.
In 1997, while reporting on HADS use, Herrmann recognized it as a reliable and valid instrument, and published studies from most medical settings worldwide provide clinically meaningful results as a psychological screening tool. By May 2002, Bjelland et al. found 747 papers that referred to HADS in Medline, ISI and PsycINFO indexed journals, and since Herrmann’s review the number of papers using HADS has increased almost fourfold. In fact, HADS continues to be one of the most useful and used instrument in different settings and conditions (Bambauer, Locke, Aupont, Mullan, & McLaughlin, 2005; Cramer, Brandenburg, & Xu, 2005; Engum, Bjoro, Mykletun, & Dahl, 2005; King Jr., Kassam, Yonas, Horowitz, & Roberts, 2005; Stoffman, Roberts, & King Jr., 2005; Vaeroy, Tanum, Bruaset, Morkrid, & Forre, 2005; Vileikyte et al., 2005; Zoger, Svedlund, & Holgers, 2004), in different age groups (Bierman, Comijs, Jonker, & Beekman, 2005; Biringer et al., 2005; Bonjardim, Gavia˜o, Pereira, & Castelo, 2005; Hedstrom, Ljungman, & von Essen, 2005; Quinlivan & Condon, 2005; Tuohy, Knussen, & Wrennall, 2005), and in different countries, languages and cultures (Barth & Martin, 2005; Engum, Mykletun, Midthjell, Holen, & Dahl, 2005; Hemmerling, Siedentopf, & Kentenich, 2005; Juang, Wang, Lu, Lee, & Fuh, 2005; Leung, Wing, Kwong, Lo, & Shum, 1999; Matsushita, Matsushima, & Maruyama, 2005; Mystakidou et al., 2005; Muluk, Oguzturk, Ekici, & Koc, 2005; Pochard et al., 2005; Skarstein, Bjelland, Dahl, Laading, & Fossa, 2005; Trappenburg et al., 2005; van der Lee et al., 2005).
The purpose of this study is to report the metric properties of the Portuguese version of the HADS.
Methods
Subjects
Participants constitute a sequential sample, including 1322 individuals, mean age 49.35 years, 60% female, with the demographic and disease characteristics shown in Table I.
Table I. Demographic and disease characteristics of the participants.
Gender Age School level Disease Number % Female M (range) M Cancer 98 100% 58.10 (25–79) 4.31 Breast (53.1%) Ovarian (10.2%) Uterine (36.7%) Stroke 253 88.1 55.26 (20–88) 6.19 Epilepsy 100 56 37.4 (14–99) 7.5 Diabetes (type 2) 316 55.4 48.38 (16–84) 6.58 Coronary heart disease 274 14.6 55.81 (22–83) 7.57 Morbid obesity (BMI, M¼48.96) 190 85.3 38.78 (15–65) 7.48 Depression 20 70 52.09 (33–68) – Myotonic dystrophy 18 44.4 42.55 (15–66) – Obstructive sleep apnoea 31 12.9 52.09 (33–68) 7.51 No disease 22 59.1 30.81 (20–60) 14.90 BMI, body mass index.
The group without disease and the group of depressed patients were used as references. For validation purposes, an analysis of variance (ANOVA) comparison between groups for age and school level was performed with Thamane’s T2 correction for not-assumed equal variances. Statistical significant differences were found for age between disease groups (F(9,1310)¼47.53, p5.0001) and for school level (F(7,1258)¼18.70, p5.0001).
Material
HADS consists of two subscales, one measuring anxiety, with seven items, and one measuring depression, with seven items, which are scored separately. Each item was answered by the patient on a 4-point (0–3) response category so the possible scores ranged from 0 to 21 for anxiety and 0 to 21 for depression. It takes 2–5 min to complete (Snaith, 2003). The HADS manual indicates that a score between 0 and 7 is ‘‘normal’’, between 8 and 10 ‘‘mild’’, between 11 and 14 ‘‘moderate’’ and between 15 and 21 ‘‘severe’’. An analysis of scores on the two subscales of a further sample, in the same clinical setting, enabled of information that a score of 0–7 for either subscale could be regarded as being in the normal range. Snaith (2003) posits a score of 11 or higher indicating probable presence (‘‘caseness’’) of a mood disorder and a score of 8–10 being just suggestive of the presence of the respective state. Bjelland et al. (2002) found cut-off points between 8 and 9. A study with a Brazilian sample (Portuguese-Brazilian language; Botega, Bio, Zomignani, Garcia & Pereira, 1995) proposed a cut-off point of 8/9 for anxiety and depression scales.
Other studies suggest that recommended cut-off scores for the HADS may result in under- reporting of psychiatric morbidity among women with early-stage breast cancer (Love, Kissane, Bloch, & Clarke, 2002).
In the Zigmund and Snaith (1983) development of HADS, authors used a sample aged between 16 and 65, but White, Leach, Sims, Atkinson, and Cottrell (1999) shows that the psychometric properties of the HADS and its conciseness make it useful in clinical settings for screenings, as well as with adolescents between 12 and 17 years of age.
A formal study of a Portuguese form is starting to be published by our group (Silva, PaisRibeiro & Cardoso, 2005).
Adaptation to Portuguese language
Translation into a foreign language demands different kinds of equivalences, such as lexical and cultural. The translation used a two-person English –Portuguese translation, a twoperson retroversion and a discussion group to achieve full consensus for the lexical and cultural equivalence. Content validity analysis was performed by two psychologists. After lexical equivalence and content validity were defined, a cognitive debriefing analysis was performed with subjects from the lower educational levels and from the oldest groups of potential patients. After the final version was defined, we chose a face format for the questionnaire identical to the one in the manual (see Appendix).
The authors recommend that caution be observed; the patient must, in fact, be literate and able to read the questionnaire (Snaith, 2003). Because older segments of the Portuguese population have lower educational levels, we were cautious about reading capabilities. The scale was generally well accepted by patients from all the disease, age and educational background groups.
Procedure
Participants are included in different research projects coordinated by the first author, sharing a common objective of the validation of HADS. They completed the questionnaire in the years 2002/2003, with the exception of the obesity group, which completed the questionnaire in 2004. All patients performed the Mini Mental Test in order to identify possible cognitive impairments. Only if they did not present possible cognitive impairments were they able to participate in the study. All participants filled in an informed consent form and satisfied all the procedures, in accordance with the Helsinki Declaration, Portuguese law and hospital rules.
Results
Considering a cut-off threshold of 11 for depression and anxiety, we identify the percentage of cases of anxiety and depression by disease presented in Table II.
Especially high numbers were found for the group of depressed patients as was expected. High numbers were found for anxiety in cancer, stroke and obesity patients. In general, the values for anxiety are higher than for depression (with exception for depressive patients) and this is the usual pattern in research (Anderson, Kaldo-Sandstro¨m, Stro¨m, & Sto¨mgren,
Table II. Percentage of subjects with a score above 11, by disease.
Anxiety (%) Depression (%) Without disease 9.1 0 Cancer 23.5 11.2 Stroke 35.1 22.5 Epilepsy 14.1 7.1 Diabetes 16.8 6.6 Coronary disease 16.4 6.2 Obesity 31.9 10.1 Sleep apnoea 9.7 4 Depression 45 55 Myotonic dystrophy 16.7 11.1
2003; Constantini et al., 1999; Johnston et al., 2000; Martin, Lewin, & Thompson, 2003; Martin & Thompson, 2000; Osborne, Elsworth, Spranglers, Oort, & Hopper, 2004). This pattern is in accordance with theory in the sense that the anxiety state seems to be a more appropriate response to the existence of a stressing disease. In contrast with the anxiety trait, which reflects the existence of stable individual differences in the tendency to respond with an anxiety state in the anticipation of threatening situations, the anxiety state is defined as an unpleasant emotional arousal in the face of threatening demands or dangers, such as having a disease. It is considered a normal part of life, and is associated with the ability to predict, prepare for and adapt to change, and creating physiological changes that enable a person to effectively respond to a threat like a diagnosis of disease.
For cancer patients, Love et al. (2002) found a different pattern, with a large amount of cancer patients in the depression group. The difference might be because the women in the Love et al. study were in the initial stages of disease and in our study they have had the disease identified for longer and therefore have experienced an adjustment period to the diagnosis.
Reliability
Cronbach’s alpha was performed to identify the internal consistency of the two scales (depression and anxiety). For anxiety, a Cronbach alpha of .76 was found with the correlation items scale corrected for overlap between .43 and .57 and with the majority of correlation in the range .50–.59. An exception is item 11 with a low correlation of .24, also found in other studies (Montezeri, Vahdaninia, Ebrahimi & Jarvandi, 2003; Muszbek et al., 2005).
For depression, a Cronbach alpha of .81 was found with the correlation item scale corrected for overlap between .37 and .64 and with the majority of correlations in range .60–.69. A correlation of .37 was found for item 8, which was also found in other studies (Botega et al., 1995). Cronbach alpha values are similar to those described in a review of international studies by Bjelland et al. (2002) from .76 to .93 for anxiety and .67 to .90 for depression.
Test –retest
In order to inspect test–retest correlations, we used two patient groups: one of diabetic patients and another of coronary heart disease patients with different time intervals, to verify if HADS is a state or trait measure.
Test–retest for 35 members of the diabetic patients group, aged between 15 and 53 years (M ¼26.14; DP ¼7.74); 40% males; 60% single and 40% married; with school levels between 6 and 18 years (M ¼12.37; DP ¼3.00), with an interval of 1 week, shows a Pearson correlation of .75 for anxiety and .75 for depression. Test–retest for 192 coronary heart disease patients, with a 3-month interval, demonstrates a correlation of .46 for anxiety and .43 for depression. These results suggest that the scales measure a state as it is expected.
Factorial validity
The manual and the authors do not refer to the factorial analysis method used to identify the factorial structure of the scale. Bjelland (2002) indicated that most studies used principal component analysis. Herrman (1997) indicated that oblique rotation was more adequate than orthogonal because it recognized an overlap between the two factors. Martin and Thompson (2000) used a maximum-likelihood factor extraction. We performed exploratory factor analysis to identify the number of factors and the magnitude of the factorial loading, using a principal component analysis
oblimin rotation method with Kaiser Normalization for the total sample of patients (n ¼1322). We found a two-factor solution explaining 46.63% of total variance; values similar to those reported by Herrmann (1997). Table III shows the structured matrix. The first is the depression factor and the second the anxiety factor as found in Herrman (1997) and Bjelland et al. (2002). Based on the factorial solution, we found discriminant values with a difference greater than 20 points between items and the factor to which they belong, versus the same items and the factor, to which they do not belong, for the majority of the items. An exception was found for item 7 in the anxiety scale, which was usual in many studies (Friedman, Samuelian, Lancrenon, Even, & Chiarelli, 2001; Herrman, 1997; Johnston et al., 2000; Mykletun, Stordal, & Dahl, 2001; Martin & Thompson, 2000), and a low loading for item 8 similarly to other studies (Flint & Rifat, 2002; Herrman, 1997; Johnston et al., 2000, Lloyd-Williams, Friedman, & Rudd, 2001). A large group of items show a great factorial weight (above .40) on the factor to which they do not belong; mainly those from the anxiety scale. The depression dimension is a more pure measure in the sense that items that constitute the scale do not load on the anxiety factor.
Confirmatory factor analysis
Confirmatory factor analysis (CFA) using EQS V6.1 (Bentler & Wu, 1995) was used to test the two- factor hypothesized model. As there was evidence of multivariate non-normality in the data (Mardia¼30.88), the Robust Maximum Likelihood estimation method was used. This has been found to control effectively for overestimation of w2, under-estimation of adjunct fit indexes and under-
identification of errors (Hu & Bentler, 1995).
The first fit index used was the Robust Comparative Fit Index (RCFI). The RCFI evaluates the adequacy of the hypothesized model in relation to the worst (independent) model. If the hypothesized model is not a significant improvement on the independent
Table III. Exploratory factor analysis.
Item
Component of structure matrix
Depression Anxiety 1 0.44 0.68 2 0.76 3 0.72 4 0.79 5 0.42 0.65 6 0.75 0.43 7 0.53 0.45 8 0.45 9 0.43 0.70
10 0.62
11 0.45
12 0.74
13 0.42 0.73
14 0.63
We display all factorial loadings above 0.40.
model, the fit indices will be close to zero (Bentler, 1995). According to Bentler and Bonett (1980), when the Goodness-of-Fit and Adjusted Goodness-of-Fit Indexes are greater than .90, the analyses indicate adequate fit of the models. Hu and Bentler (1999) refers values of .95 or higher. The second fit index was the Root Mean Square Error of Approximation (RMSEA; Steiger, 1990) where a value of .05 or lower indicates a good fit and values up to .08 indicate an acceptable fit (Browne and Cudeck, 1993). The RMSEA has been described as ‘‘one of the most informative criteria in structural equation modeling’’ (Byrne, 1989, p. 112.)
In our study, CFA results for the two-factor solution, w2(76)¼334.90; RCFI¼.95; RMSEA¼.05 (95%
confidence interval¼.046–.057) showed that fit indices for the twofactor solution are good. These values are similar to those found by Martin et al. (2003).
Comparison between groups
A comparison between disease and non-disease groups was performed with ANOVA.
In simple one-way between-subjects ANOVA, problems created by unequal group sizes are relatively minor. However, as group sizes become more discrepant, the assumption of homogeneity of variance is more important, explains Tabachnick and Fidell (1996). These authors explain that in non-experimental work unequal n often reflects the nature of the population. Inspection of homogeneity of variances with the Levene test examines the assumption that variance of the samples is homogeneous. For anxiety, the assumption is accepted, but not for depression (p ¼.001).
An ANOVA comparison for anxiety between groups was performed. Statistically significant differences were found for anxiety between disease groups (F(9,1303)¼9.83, p5.0001). An ANOVA comparison for depression between groups was performed with Thamane’s T2 correction for variances not assumed equal. Statistically significant differences were found for depression between disease groups (F(9,1304)¼24.55, p5.0001). Table IV shows the means for anxiety and depression by disease groups.
Results for depression show higher mean values for depressive patients, followed by stroke and obesity patient groups, and low mean values for the no-disease, obstructive sleep apnoea and epilepsy patient groups. Results show higher mean levels of anxiety for depression, cancer and obesity patient groups and low mean levels for obstructive sleep apnoea, myotonic dystrophy patients and no disease groups. Values for depression are lower than
Table IV. Anxiety and depression means by disease group. Anxiety
M
Depression
Disease group Disease group M No disease 7.81 No disease 3.22
Cancer 9.18 Cancer 5.89
Stroke 10.53 Stroke 9.18
Epilepsy 8.59 Epilepsy 5.18
Diabetes 8.18 Diabetes 5.46
Cardiac coronary disease 8.41 Cardiac coronary disease 5.53
Obesity 10.52 Obesity 7.28
Obstructive sleep apnoea 6.87 Obstructive sleep apnoea 3.83 Depression 11.00 Depression 12.90 Myotonic dystrophy 7.27 Myotonic dystrophy 7.22
levels for anxiety in all groups except for depressive patients; these results match the content validity of the scale.
Correlation between scales
The correlation between anxiety and depression is, in general, high. Lovibond and Lovibond (1995) explain that depression and anxiety are quite distinct, but the clinical overlap between the two conditions has long attracted both clinicians and researchers with self-report anxiety and depression scales typically correlated between .40 and .70 across a wide range of patient and non-patient samples. Anxiety scales frequently correlate as highly with depression scales as with other anxiety scales, and depression scales show an equal lack of specificity. This is an important issue in the distinction of the two constructs, which is why it is important to identify the correlation between the two constructs under investigation.
The correlation between depression and anxiety scales of the HADS in the present study shows an r(1307) ¼.58, p5.0001. It is a moderately high correlation, but lower than those reported by Herrmann (1997), and similar to the values reported by Bjelland et al. (2002). It is high enough to raise the question: can HADS still be used as a two-dimensional instrument?
Discussion
When someone has a diagnosis of disease, it is a normal reaction to experience changes in emotional state. If the emotional reaction, namely anxiety and depression, is high it can have a negative impact in the adjustment process and in the disease process (confuses the diagnosis, prolongs the recovery time, treatment adherence). The identification of the magnitude of the emotional reaction becomes important to organize the support to the patient. Snaith (2003) defend that there can be no doubt of